← Predictions

Recorded before kickoff

The forecast. The timestamp. The result.

Explore retained forecast versions and see how their probabilities performed after the final whistle.

Browse forecasts ↓ See the full track record →

Original probabilities stay unchanged. Results are graded separately. The archive starts when retention was enabled; earlier forecasts are not reconstructed.

Public calibration - frozen forecasts

Latest 500 completed fixtures with retained forecasts; one latest retained pre-match version per fixture. Includes all model versions shown below. No reconstructed forecasts enter this report.

494graded matches

0.5506Multiclass Brier (0–2)

0.9321Log loss

500 considered · 0 awaiting results · 6 excluded: provenance or result could not be verified

Show the calibration table

Calibration across home, draw and away outcomes
Probability band	Mean predicted	Observed frequency	Outcome observations
0–10%	6.1%	7.2%	69
10–20%	15.8%	14.3%	251
20–30%	24.8%	22.6%	523
30–40%	34.9%	35.0%	200
40–50%	45.0%	50.3%	171
50–60%	54.7%	57.0%	107
60–70%	64.3%	65.4%	81
70–80%	74.8%	82.4%	51
80–90%	84.4%	73.9%	23
90–100%	94.0%	100.0%	6

Each match contributes three outcome observations; these are correlated, not three independent matches. Bands include the lower bound and exclude the upper bound, except 100%. Sparse bins are unstable. Lower Brier and log loss are better.

Model versions: v5 (494)

Updated 14 Nov 2026 · 22:40 America/New_York · Forecasts stay unchanged. Grading uses current corrected regulation results; extra time and penalties are excluded. Rounded probability vectors are normalized for scoring; log loss uses natural logs with a 1e-15 floor.

The evidence

Published forecasts

Times shown in America/New_York. The highlighted probability is the published pick; the underlined outcome is what happened after 90 minutes.